Improving windows tasks recognizer for Assamese using bigram analysis

Assamese is the official language of Assam, a state located in the North Eastern part of India. It is spoken all over Assam and some parts of its neighboring states. Being the host of a number of communities, the language in its recent form is a mixture of vocabulary from many languages with many pronunciation varieties. A lot of research work has been done to know the various aspects of this language. However, not much work has been done to connect recent technologies not only with Assamese but with other North East Indian languages. This paper presents the research work done in the area of Assamese phonetics to design a voice driven system for controlling Windows tasks. Phonetic transcription was done for about five thousand Assamese words, some nonsense words along with commonly used numbers. The result of the analysis is described in the form of bigram tables of vowel-vowel and consonant-vowel sequences in the analyzed set of words. It also describes the implementation of a voice recognizer for several Assamese words that has been extended to an application controlling the basic Windows tasks via voice commands. The Hidden Markov Model Toolkit was used to deploy the system. Triphonebased HMMs were constructed and used to run the recognizer with live input. One Win32 application was designed on top of this recognizer that uses the functions provided by Win32 APIs to implement some of the Windows tasks, once the word or set of words are recognized. The designed application is capable of running several Windows applications using Assamese voice commands which enable persons not familiar with English to use the system. This work also substantiates the fact that phonetic analysis and embedding the characteristics of a language is crucial to developing high quality speech recognition systems.